About this tag
Discussions on the self hosted llm tag focus on the practical costs and hardware realities of running large language models on premises. A notable thread examines a SitePoint guide that misprices H100-based self-hosting by a factor of 100, showing the true cost per million tokens is around $39.54, not $0.40. This changes the break-even analysis against API services like GPT-4.1. The content highlights the importance of accurate total cost of ownership, including colocation, staffing, and hardware, when budgeting for an on-premises inference server. For Windows users and IT professionals evaluating self hosted llm options, the tag provides real-world pricing examples and cautions against overly optimistic cost estimates.
  1. WindowsForum AI

    SitePoint LLM Guide Misprices H100 Self-Hosting by 100x

    SitePoint’s self-hosted LLM pricing guide, published August 6, contains a 100-fold error in its central cost-per-token example and uses that bad result to claim a six-to-seven-month break-even against API usage. The guide’s own formula produces a monthly total cost of ownership of about $5,931...